May 24, 2020

Overview of Compartment Syndrome

What is Compartment Syndrome?

Why does Compartment Syndrome occur?

How is Compartment Syndrome treated?

Review of Compartment Syndrome & Fasciotomy

In this study, we will use Fasciotomy as the indicator for Compartment Syndrome.

  • We did this for the following reasons:
    • Compartment Syndrome is a diagnosis that is subject to human error
      • The PTOS results are sparse in terms of date, time, and accuracy of diagnosis
    • Fasciotomy is a procedure that takes place because of Compartment Syndrome
      • This is a surgical procedure and it is well documented in the PTOS data

Overview of Forearm Factures

Four Types of Forearm Fracture Treatments

  1. Open Reduction with Internal Fixation
    • Most common and preferred method for forearm fracture treatment
  2. Closed Reduction without Internal Fixation
    • Involves manually setting the bone back into place and use of hard cast for stability
  3. Closed Reduction with Internal Fixation
    • Not a surgical procedure, minor incision in forearm using pins for stability
  4. Open Reduction without Internal Fixation
    • Surgical procedure, only performed if there is no need for internal fixator

Review of Open Reduction Forearm Fractures

Review of Closed Reduction Forearm Fractures

Review of the Four Forearm Fracture Treatments


  • These are surgical procedures
  • More traumatic fractures requiring invasive techniques
  • Use of plates, screws, and external cast to secure fracture site
  • Patients are sedated using anesthesia
  • These are not surgical procedures
  • Technique performed by manipulation for less traumatic fractures
  • Use of pins and cast to hold bones in place for healing
  • Painful technique for patient and performed with or without general anesthetic

Pennsylvania Trauma Outcome System

What is Pennsylvania Trauma Outcome System?


This is the primary data source used for our analysis of Compartment Syndrome in Pediatrics Patients.

  • Approximately 530k anonymized patient records and 1,400 columns of descriptive information regarding each procedure
  • Medical results occurring between 2010 - 2015
  • Patient demographics, diagnosis, prehospital procedures, surgical procedures, and outcomes
  • Uses International Statistical Classification of Diseases and Related Health Problems (ICD) medical coding to classify causes of injury, examination results, processes, and treatments.


Pennsylvania Trauma Outcome System will be referred to as PTOS going forward.

Objective of Compartment Syndrome Analysis

Objective of Analysis

Identify the likelihood of compartment syndrome in pediatric patients who suffered from forearm fractures .


Data Preparation

Data Preparation

  • Developed a custom R package, called traumaR to translate all medical codes into human friendly terms
    • Wrote over 50 functions to automate this work and make it repeatable
  • PTOS Data consists of 530k anonymized patient records and 1,400 columns
    • Columns mostly consisting of medical codes which required translation
    • Necessary to normalize the data into structured data frame
  • The final results create two main data frames joined together by patient_id
    • Patient Information
    • Medical Procedures

traumaR

Exploratory Data Analysis

EDA: Patient

EDA: Forearm Fractures

EDA: Fasciotomy

EDA: Fasciotomy Results

The Open Reduction with Internal Fixation is the most common technique across the medical field

EDA: Fasciotomy Results



  • Pediatric patients tend to be treated with Closed Reduction and casting
    • Most of the injuries result from falling on the playground or at home
    • These fractures are usually less traumatic and do not require invasive techniques

Association Analysis of Medical Procedures

  • The red lines represent common paths in the series of procedures that led to a fasciotomy.
  • Removed CAT scans (87.03, 87.41, 87.71, 88.01, 88.38), and suture code (83.65) procedures as they made up a majority of the relationships.
    • The number of CAT scans were related to the severity of the injury and suturing is just part of the everyday surgery.

Modeling

Predicting Fasciotomy

Model Pre-Processing

  • Up-sampled the minority class in training data
    • Fasciotomy == TRUE
  • Use F-Measure as measure of model accuracy
    • F-Measure combines precision and recall into a single score
    • Highly recommended for imbalanced classifications

Feature Engineering

  • Step Other
    • Groups categorical information into ‘other’ category whose individual levels do not exceed preset threshold.
  • Step Dummy
    • Converts character or factors into dummy variables
  • Step Zero Variance
    • Removes variables that contain only a single value
      • This is a precautionary step to ensure that Step Other captured all low occurrence into ‘Other’
  • Step Omit NA
    • Remove any observations with missing values
    • Some models cannot handle missing values

Models Deployed


Model Results: In-Sample Accuracy

Model Results: Out-of-Sample Accuracy

RandomForest Variable Importance


Variable Importance

  • Specific Types of Injuries
    • Fall from Stairs, Motor Vehicles, Motorcyclist, Firearm Explosives
  • Location of Injury
    • Home, Street/Highway
  • Patient Demographics
    • Male, White, and Age 22
  • Medical Procedure
    • Operations on Skin and Subcutaneous Tissue

ROC Curve

Train Data

Test Data

Area Under Curve Scores

Visualization and Application

Shiny Dashboard: Web Version

Shiny Dashboard: Phone Application

Review of Deliverables & Output

Review of Deliverables & Output

Review of Objectives

  • Created a model that can predict the likelihood of a patient developing compartment syndrome based on patient demographics and medical history.
  • Developed a shiny application to assess likelihood of compartment syndrome with user-friendly interactive interface.
  • Built a customer R Package, traumaR, and published it to GitHub and made it publicly accessible.
  • Contributed to another R Package, tidytable, during the project which is published on CRAN.

Reflection on Project

Reflection

  • Overall, the results showed that severity of injury and cause of injury were the leading predictors of compartment syndrome.
  • There was little information presented to confirmed that the order of medical procedures had any part of increasing compartment syndrome.
  • The PTOS data provides significant insights for the medical field; however its initial data structure invites challenges for individuals lacking data manipulation expertise.